Sean Reilly

Building a Personal AI Infrastructure in 28 Days

Fri, 23 Jan 2026 12:00:00 +0000

Most people treat AI conversations as disposable. Ask a question, get an answer, close the tab, forget the context. I did that for two years and despite pruning ended up with 1,038 ChatGPT conversations - a graveyard of half-finished thinking, abandoned research, and strategic planning that existed only in a chat history I’d never revisit.

The problem isn’t the AI. The problem is that every conversation starts from zero. Your AI has no memory of what you figured out last week. No access to the decisions you made last month. No understanding of the context you’ve spent years building. Each session, you’re re-explaining your business, your preferences, your constraints, and then throwing away whatever you built together.

I decided to fix that. Over 28 days, I built a Personal AI Infrastructure (PAI) that actually remembers what I’ve learned.

The Uncomfortable Truth

ChatGPT was first with memory, Claude was first with projects; their features have improved significantly, but custom instructions are still limited to 1,500 characters.

When OpenAI decided to change how Projects worked, users had to manually migrate their instructions or lose them (like I did). ChatGPT also had a catastrophic failure in February 2025 that lost users’ conversation history, and Claude Projects has a known bug right now where it forgets uploaded file contents.

I realised that my carefully crafted context, refined over months, existed at the discretion of these vendors. It’s all subject to change, and not truly portable.

Anyone who’s worked with me knows I have trust issues around vendor dependency. I’ve seen too many teams tied to a terrible service provider, trapped when a platform changes pricing, changes a feature you relied on, gets acquired, or just disappears. I’ve watched organisations lose years of institutional knowledge because it lived in a tool they can no longer use.

Ask me about the time when working in WeWork, Soho, when I went for a quick coffee, bumped into a Shopify solutions architect, and ended up arguing this very point with him for over over an hour and a half.

Do you want your complete business knowledge architecture locked in a platform with that level of uncertainty? Or today’s essential tool becomes tomorrow’s ad-supported platform.

OpenAI recently announced they’re exploring advertising because subscription revenue isn’t enough. Their financial position is precarious - losses reportedly outpacing revenue three-to-one, with over a trillion dollars in planned infrastructure investment contingent on achieving Artificial General Intelligence (AGI).

Introducing SAI

If Claude, OpenAI, or any other tool vanishes tomorrow, I wanted to be in a position where I only lose a tool, not years of accumulated thinking.

Over Christmas 2025, I built what I call SAI. SAI stands for Sean’s AI (obviously).

SAI is my Personal AI Infrastructure, that actually remembers what I’ve learned and removes the dependency and vendor lock in that I so despise. I liked the double meaning of SAI: in Sanskrit, “sai” means divine or master; in Japanese (才), it means talent or ability.

Will SAI give me God-like talent? I’m working on it.

The Core Idea: Filesystem as Brain

The principle is simple: all permanent knowledge lives in files, not chat histories.

Every AI conversation that produces something worth keeping gets captured in a markdown file. Every decision, every piece of research, every strategic plan goes into a folder structure I control, tracked by Git, and accessible to any AI tool I choose to use.

When I start a new conversation with Claude, ChatGPT or a local AI model (Mistral, LLama, DeepSeek) I can point it at the relevant files. The model reads my previous thinking, understands the context, and picks up where I left off. No re-explaining, context loss or starting from zero.

Git commits become a memory system - 204 commits over 28 days tracking how my thinking evolved, what decisions I made and why, what I tried that didn’t work.

What I Actually Built

The technical setup breaks down into a few key pieces:

Multi-AI routing: I use different AI tools for different jobs. Claude Code for architecture and multi-file work. Gemini for research and current information. Codex for quick terminal queries. Ollama for anything involving sensitive data that shouldn’t leave my network. The AIs can even delegate to each other - Claude can spin up a Gemini research task when it needs current information, then continue once the results come back.

Context files: Standardised markdown files that any AI can read. Project instructions, decision matrices, handoff protocols. Each project gets its own context file, so I can isolate context to only that project when needed. When I switch between tools or come back after a break, the context travels with me.

Automation: Shell scripts and aliases that reduce friction. A command to initialise a new session. A script to hand off work between AI tools. Integration with Todoist and Notion via MCP servers.

Storage: A ZFS dataset on my server, mounted to my AI workstation. ZFS gives me snapshots (I can roll back if something goes wrong), compression (saves space), and checksums (catches corruption before it causes problems).

Honestly, the MCP server setup was harder than expected. OAuth troubleshooting ate most of a day, and the Todoist MCP server is still very buggy - it works, but not perfectly. But once it worked, having my task manager and notes system directly accessible to SAI changed how I work.

How I’m Using It

I’m using SAI for career planning and a project I’m calling Harbour. I won’t go into the details here - it’s the start of a ten-year plan to make some significant life changes - but the point is that this system works for strategic thinking, not just technical tasks.

SAI can access any source, and I can use the strengths of any model to access context across weeks of thinking; to have SAI remember what I was considering, what I ruled out and why, and what the constraints are.

I’ve used SAI to unify eleven years of health data from multiple wearables and apps - Jawbone, Garmin, Withings, Sleep as Android, Google Fit - into a single analytical layer. Over 2,500 weight records, 2,800 sleep records, and 3,300 daily step counts, all normalised and queryable.

What initially looked like bad data to the model turned out to accurately reflect my hybrid working patterns, followed by COVID-enforced lockdowns between 2018 and 2021. I also discovered that Garmin was over-reporting my sleep by several hours each night when compared with Sleep as Android data, and that my sleep quality had actually been declining since mid-2024 - something I hadn’t spotted in any single app.

These patterns only emerged from analysis of unified, long-term data.

Business Use Cases

A PAI isn’t just a personal productivity hack. The same approach would scale for an organisation as an EAI (Enterprise AI Infrastructure):

Marketing teams Imagine a new marketing director who inherits years of campaign learnings, competitor analysis, and strategic decisions that are all searchable, and with all context intact. No more institutional knowledge walks out the door; no more “Sorry, the person who knew that left”.

Executive strategic planning compounds over time. Each planning session builds on the last one as the EAI remembers what you tried, what worked, and what the constraints were.

Cross-functional knowledge including product decisions, customer research, and technical architecture are captured by the EAI in a format that survives staff turnover and tool migrations.

Compliance and audit trails become straightforward; every decision is documented with full context by the EAI and committed to Git so that when someone asks “Why did we do it that way?”, the answer exists and is findable.

Information Security is in your control with a local model, as PII or privileged financial data stays within your EAI rather than being parsed by third-parties.

You Don’t Need Enterprise Infrastructure

I built this on Atlas, using my repurposed HP workstation which I use as a Proxmox server.

The same approach would work on:

A Mac with Homebrew and a folder
A Windows PC with WSL2 (or even just PowerShell)
Even a Raspberry Pi, for the truly committed

The filesystem-as-brain principle works anywhere you can run Git and a text editor. The infrastructure I used made certain things easier, but the core idea is platform-agnostic by design.

I prefer open source where possible. Ollama for local models, Fabric for prompt patterns, ZFS for storage, Proxmox for virtualisation. But I’m also realistic - Claude and OpenAI have vastly more compute power than my home server. The solution is to use cloud AI for heavy lifting, but keep all knowledge in files I own. Even if the internet goes down, I can still work with a local model.

The Numbers

I need to be honest here, I invested time and money upfront to get SAI going using top-tier Claude MAX, which is £150/month.

28 days from zero to production:

204 commits tracking evolution of thinking
1,038 ChatGPT conversations migrated and organised
2,595 conversation files standardised to consistent naming
4,469 lines of AI context and instructions across the workspace
233+ AI prompt patterns available via Fabric
3 MCP servers connected (Todoist, Notion, n8n)
£25-75/month ongoing cost (Claude subscription plus minimal API usage)

Credit Where It’s Due

This is where I owe a huge thank you to Daniel Miessler, whose PAI concept gave me the foundation for this entire project. What I’ve built is my implementation of his ideas, adapted to my specific needs and workflow.

If you want the richer, more detailed version his work, this is where to start:

For a gentler introduction, NetworkChuck’s AI in Terminal video gets you running in an afternoon - though you’ll need coffee (don’t worry, he’ll remind you), because Chuck’s energy level is… a lot. Daniel’s approach is more powerful and comprehensive, but Chuck’s is more accessible if you’re not sure where to begin.

Inspiration:

Daniel Miessler - PAI concept and Fabric
NetworkChuck - Accessible AI terminal intro

Tools I used:

Claude Code - Primary AI interface
Gemini CLI - Research and web search
Fabric - AI prompt patterns
Ollama - Local AI models
Proxmox - Virtualisation platform
ZFS - Storage filesystem

What’s Next

The system is working. The knowledge compounds. If any of these tools disappear tomorrow, my thinking stays with me in files I own, on storage I control, backed up in ways I chose.

But the current setup requires me to be at my desk, SSH’d into my AI workstation. That’s fine for deep work, but limiting for quick queries on the move. Mobile SSH clients exist, but they’re fiddly - connections drop, typing on a phone keyboard is painful, and complex commands become error-prone.

I’m building an integration with n8n and Slack to enable mobile access (thanks again to NetworkChuck for the inspiration).

The goal is a Slack bot that can reach SAI, run queries, keep conversation threads organised, and move files back and forth. This would be more connection-robust and less typo-prone than my current mobile solution Termux, which disconnected three times while I was working through this post on the train.

If you’re happy with throwaway AI conversations and don’t mind re-establishing context each session, the PAI approach is probably overkill. If you’re not comfortable with command-line tools and Git, the learning curve might not be worth it.

But if you find yourself frustrated by context loss - if you’ve ever thought “I figured this out before, why am I explaining it again” - this approach is worth exploring.

For a organisation, the EAI approach could be transformational. Institutional knowledge that survives staff turnover. Strategic context that compounds instead of resetting. AI assistants that actually understand your history, constraints, and decisions. The infrastructure investment is modest; the returns compound indefinitely.

Start with Daniel Miessler’s work. Watch the video. Read his writing on fabric. You don’t need to build exactly what I built. The principle - filesystem as brain, knowledge in files you control - is what matters.

AI Search & SEO - What Marketers Should Really Be Asking

Mon, 30 Jun 2025 08:00:00 +0000

When someone claims, “I searched this in ChatGPT and we didn’t show up,” it’s tempting to shrug it off. But in the messy world of AI-powered search, those offhand comments can spark legitimate concerns, and they need a structured response.

I put together a checklist after a board member raised exactly this question. When I emailed it to colleagues, it saved a lot of back-and-forth - so here’s a version for anyone facing similar conversations.

Why Asking the Right Questions Matters

Perspective check: AI chat tools make up only around 3% of total web search traffic. Google still accounts for over 90% market share globally. AI search is growing fast and worth watching, but it's nowhere near replacing traditional search.

AI chat-based tools like ChatGPT, Perplexity, Gemini, and Copilot are growing fast in visibility. That means marketing should still prioritise traditional SEO - AI chat tools aren’t replacing Google or Bing anytime soon.

Still, someone asking “why don’t we appear in ChatGPT?” deserves a decent response. The same phrasing or tool can produce very different results depending on a few key variables:

Prompt wording
Follow-up queries
Model version and training cut-off
Any stored memory or past chat sessions
Plugins, retrieval tools, or browsing enabled

So instead of guessing, we ask the right questions.

Questions to Ask When Auditing an AI Search Claim

Here’s a quick checklist to get the right context before we jump to conclusions:

Which AI tool did they use?
ChatGPT, Claude, Perplexity, Gemini, Harvey… If they mention self-hosted tools (e.g. Ollama, DeepSeek) or say things like “13b model,” tell them to keep reading - I explain it below.
What model version was used?
GPT-4, GPT-4o, Claude 3.5, Gemini 1.5, etc.
What was the initial prompt?
The first thing they typed or said. Get a copy if possible.
Were there follow-up questions?
AI is conversational. One follow-up can change the whole thread.
Was there any stored memory or chat context?
If they’re not sure, they can ask the AI:

“What context or memory are you using for this answer?”
Were there any system settings or preferences active?
Things like tone of voice, safe-search filters, or restricted sources.
Was it a fine-tuned or company-trained model? If yes, that’s a different game entirely - and probably beyond what most marketers need to worry about.
Can they share the full transcript or a chat link?
Most paid tools (like ChatGPT Plus) allow sharing.

Model Size ≠ Output Quality

Marketers sometimes hear terms like “7b,” “13b,” or “70b” and assume bigger = better. That’s not quite right.

Those numbers refer to the number of parameters in a model: the internal “knobs” it uses to understand language. A 70b model has more of these than a 13b one. But that doesn’t guarantee better answers.

Think of it like engine size: more power doesn’t always mean a better ride. My son James is a talented racing driver, and I’ve watched him hustle a Citroen C1 around Silverstone against BMW Z4s, Mazda RX8s, and Honda Type Rs. It’s about training, tuning, and context.

Why This All Matters

Google has begun surfacing Gemini activity in Search Console, but only for AI Overview impressions. I was at Google’s offices in June for the Amadeus Google Summit and spoke directly with a Senior Product Manager about this - it’s early days, and far from comprehensive.
There are now a few unofficial tools that let you peek into Gemini or ChatGPT behaviour. I spoke with a large SEO agency who’d just built and launched one - they were honest that it’s useful for investigation but fragile, manually operated, and nowhere near scalable.
Crucially: AI models don’t retrieve results like search engines. They generate responses based on data, prompts, and model behaviour - not a ranked list of pages.

Bottom Line for Marketers

Traditional SEO still matters most. It powers both human and AI understanding.
Treat AI visibility checks as qualitative research, not a KPI.
If someone flags something from an AI chat, ask for the right context before jumping in.
And if they say “7b” like it’s gospel, smile and tell them to come talk to me.

Professional video and audio for remote working

Mon, 20 May 2024 11:11:00 +0000

A Complete Guide to Setting Up Your Home Studio for Professional Video Recording

In today’s digital age, appearing professional on video calls and recordings is essential. Whether you’re creating content, conducting business meetings, or simply want to look your best on Zoom, having the right equipment can make a significant difference. While I don’t publish videos of myself, I use podcast/streaming-quality equipment for my Zoom calls, which helps me appear more professional than using a laptop’s built-in microphone and camera. Here’s a detailed look at my setup, balancing cost and quality.

My Desk Setup

Here’s a snapshot of my desk setup, featuring most of the equipment discussed below:

My home studio: Logitech Blue Yeti mic, Brio 4K camera, and dual Litra Glow lights.

Why Audio Matters Most

Audio quality matters more than video. Poor audio will undermine even the best camera setup. Viewers will tolerate grainy video, but they'll click away from bad sound.

Before diving into the equipment, it’s crucial to understand the importance of audio. Poor audio quality can detract from your message, no matter how good your video looks. To understand why audio is so critical, check out this video.

Microphone: Logitech Blue Yeti

A high-quality microphone is essential for clear audio. I use the Logitech Blue Yeti, which is highly praised for podcasting. It’s available in several colors if you want it to be seen on camera:

The Blue Yeti offers excellent performance for its price point. For setup tips and reviews, check out these videos:

I use a custom 3D-printed microphone base to keep it out of the way of my screens but directly in front of me. Alternatively, you can mount it on a microphone arm like this one. You might also want a pop filter if the mic is close to you; here are some options.

Camera: Logitech Brio 4K

For video, I use the Logitech Brio 4K streaming camera, which offers great quality for its price. It has become more affordable recently due to the release of a newer model (the MX):

If you have a DSLR that allows streaming, this can provide even better quality with manual focus capabilities. For more on using a DSLR as a webcam, check out this guide.

Lighting: Logitech Litra Glow

Good lighting is crucial for professional-looking videos. I use two Logitech Litra Glow lights on both monitors to ensure consistent lighting without casting shadows on my face:

Logitech Litra Glow

For review and setup ideas, watch this video.

Software: OBS (Open Broadcaster Software)

All of this equipment feeds into OBS (Open Broadcaster Software), a free open-source software I use to crop the view around me, hiding unwanted items and adding my logo to the feed. OBS can record or stream to platforms like Zoom.

Outdoor Recording

For outdoor recording, sound remains the priority. Here are some options:

Wired Lavalier Mic: The Boya BY-M1 is a cost-effective choice, though it limits your range.
Wireless Mic: The RØDE Wireless GO II offers more freedom, allowing recording up to 100 meters away from the camera.

Check out footage I shot using the Rode mic here. Note: Avoid the white version as it quickly becomes grubby.

Handheld Outdoor Recording

For handheld outdoor recording, a good gimbal is essential to keep your footage steady. I use the Hohem iSteady M6 with a tracking module:

Check out various bits of footage shot with the iSteady:

Setting up a professional-quality recording setup at home doesn’t have to break the bank. With the right equipment and a bit of know-how, you can significantly enhance the quality of your videos, whether for business or personal use.

How to Disable Caps Lock Notifications in Windows 11

Tue, 28 Feb 2023 10:05:00 +0000

Nothing ruins a tense moment in a first-person shooter quite like accidentally hitting Caps Lock and having Windows throw a notification across your screen. This happened to me yesterday, mid-firefight, and I finally snapped and fixed it properly.

If you’re here, you probably have the same problem. Here’s how to turn it off.

Windows 11 Settings

This fixes the system-level notification and the toggle key sounds:

Open Settings (Win + I)
Go to Accessibility > Keyboard
Turn off Toggle keys (this stops the sounds when you press Caps Lock, Num Lock, or Scroll Lock)

That’s it for most people. But if you have a Dell machine, there’s another culprit.

Dell Peripheral Manager

Dell installs its own software that adds a separate Caps Lock indicator. I’d forgotten this was even on my machine until the Windows fix didn’t solve the problem.

Open Dell Peripheral Manager (search for it in Start)
Select the Keyboard tab
Toggle off Caps Lock Indicator

If you don’t have Dell Peripheral Manager installed, you don’t need to worry about this step.

Why I Keep Finding My Own Blog Post

I’ve fixed this at least three times now on different machines and after Windows updates. Each time I google it, find this post, and remember I already solved it. If you’re future me: hello. If you’re someone else with the same problem: hope this helped.

Useful Digital Marketing Learning Resources

Mon, 01 Jan 2018 17:51:57 +0000

Thought I’d share this with everyone just in case there are any nuggets that might be useful for others.

Tool Blogs

Any of the blogs for the tools that are used can be great, especially Moz and their Whiteboard Friday videos:

Moz Blog

Also the following tools - bearing in mind their content will skew towards using their tools but can be widely applied:

Official Sources

The Official Google Webmaster Tools Blog is the source for many new updates:

Google Webmaster Central Blog

Many official updates, plus things that are being widely noticed by the community but not being talked about officially by the search engines, will come out on these closely related sites:

Events

If you can make it up to town, Search London is worth attending. They have regular speakers such as Aleyda Solis, Lukas Zelezney, Russell McAthy, etc. as well as the founders/directors of many of the other agencies and toolsets. Good opportunity to network, normally paid ticket but you get a load of content and normally a couple of free drinks:

Search London Meetup

Also Brighton SEO - tickets can be sourced for free if you are lucky, as well as paid for with additional benefits such as premium sessions and round tables. If you can’t make it, stream the main session live for free:

Brighton SEO

Look for the regular authors/speakers on any of the above, especially guest authors, and check out their twitter feeds for content they write or share.

Courses

If you want to get a wide overview on all digital marketing topics, spend £11 on the following course on Udemy - 19 hour course with video and specific tasks to do:

Udemy Digital Marketing Course

That should get you started. I haven’t touched outreach, social or local there (saying that Greg Gifford is good for local SEO albeit with a US skew - definitely watch his content, it’s brilliant).

Broader Marketing

Finally, look to the wider marketing field as well, especially as digital marketing becomes more just marketing. Marketing Week is good in this respect and especially the content from Mark Ritson who is a very straight shooter (plus adjunct professor of marketing for Sydney University and marketing advisor for LVMH) - he is not a social media marketing fan though!

Mark Ritson on Marketing Week

Large Ecommerce Sitemap Specifications

Sun, 28 May 2017 17:51:57 +0000

A sitemap specification I wrote in 2017 for a large ecommerce client with significant inventory churn. The development manager requested the Agile Given/When/Then format. Published here for reference.

Given that we have assets on the websites and that we want search engines to find them and index them quickly
When the assets are published
Then they should be listed in a XML sitemap file in order to make it quick for the search engines to find

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And these assets are of different types
And some of them are webpages
And some of them are images
And some of them are videos
When the assets are published
Then they can be grouped together where these assets all feature on one page
And we need to group related items together within <url> </url> tags
And we need to encapsulate the items in the correct tags for their type

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And that they belong to a variety of different categories
And that some of these categories have specific sitemap formats
When the assets are published and added to an XML sitemap
Then we need to make sure that these sitemaps are split into various categories
And each category has its own file
And if a category has its own format requirement that we use that format

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we wish to sell our products
And we have a large amount of stock
And this stock is categorised by different brands
When the assets are published and added to an XML sitemap
Then we need to list products of a given brand in their own sitemap xml
And this should be called Brand.xml so it is quick for humans to understand what is in it (for example Category.xml)
And this can contain the URL groups that include pages, images and videos
And this sitemaps should contain the URLs for the root page for the brand (e.g. https://www.website.co.uk/Category)
And this sitemaps should contain the URLs for the root page for all series under this brand (e.g. https://www.website.co.uk/Category/SubCategory)
And where applicable this should contain the URLs for the root page for models and sub models where these pages exists (e.g. https://www.website.co.uk/Category/SubCategory/16618, https://www.website.co.uk/Category/SubCategory/116610-LV, https://www.website.co.uk/Category/SubCategory/116610-LN)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we wish to buy products from the general public
And this stock is categorised by different brands
When the assets are published and added to an XML sitemap
Then we need to list products of a given brand in their own sitemap xml
And this should be called SellBrand.xml
And this can contain the URL groups that include page URLs only (not images or video)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we have a New Arrivals section drives many new sales
And it would benefit us to make this most regularly crawled section of the website by search engines
And search engines tailor their crawl frequencies based on files they see changing rapidly
When the assets are published and added to an XML sitemap
Then we need to list the latest 100 new arrivals page URLs on their own NewArrivals.xml sitemap
And new arrivals that are older than the latest 100 need to be removed
And this sitemaps does need to contain URLs but does not need to list image or video details on it
But this sitemap can and will list URLs that are on product sitemaps (Category.xml etc.)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of these pages are relating to our service centre
When the assets are published and added to an XML sitemap
Then we need to list all servicing pages in a Servicing.xml sitemap
And this does need to contain images and videos

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of these assets are articles
And some of these assets are news
When the assets are published and added to an XML sitemap
Then we need to list all the pages that are in this category in a Articles.xml sitemap
And this should list all images and videos that relate to the articles
And we should also add them to specific News.xml for the first 48 hours that they are published
But then remove them after 48 hours from the News.xml

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of them do not fit into the categories listed (homepage, info pages, about us, awards, promos/sale pages etc.)
And that we do want these to be picked up by the search engines
When the assets are published and added to an XML sitemap
Then we need to list these in a mop up sitemap Website.xml

Given that we have assets on the websites and that we want search engines to find them and index them quickly
When the assets are published and listed in an XML sitemap
Then we need to make sure the XML file is Encoded correctly
And we need to tell the search engines it is so by starting each file with <?xml version="1.0" encoding="UTF-8"?>
And we need to tell search engines what types of URL sets we are publishing

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we need to tell search engines what types of URL sets to publish
And where they can find the matching schemas
And that we have pages, images and videos
When the assets are published and added to an XML sitemap
Then we need to list these URL schema sets within the <urlset > tag
And for pages the schema is xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
And for video the schema is xmlns:video="http://www.google.com/schemas/sitemap-video/1.1"
And for images the schema is xmlns:image="http://www.google.com/schemas/sitemap-image/1.1"
And it should look like this <urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9" xmlns:image="http://www.google.com/schemas/sitemap-image/1.1" xmlns:video="https://www.google.com/schemas/sitemap-video/1.1">

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of them are web pages
When the assets are published and added to an XML sitemap
Then we need to list the locations of URLs using <loc> </loc> tags
And these need to be within the URL group which is wrapped <URL> </URL> tags

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of them are images
And we can provide a title which will help the search engines understand what is in an image
When the assets are published and added to an XML sitemap
Then we need to list details of images using <image:image> </image:image> tags
And we need to list the URL of an image within <image:loc> </image:/loc> tags
And we need to provide a title in <image:title> </image:title> tags
And this image, where it is a product should contain the H1 of the product page, the angle of the shot and the product code
And we need to provide license details which will be for every one <image:license>[https://www.website.co.uk/info/Licensed-Images</image:license>

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And some of them are videos
And we can provide additional details about these videos
And we host all our videos on YouTube
When the assets are published and added to an XML sitemap
Then we need to list details of videos using <video:video> </video:video> tags within the URL group below the page location details
And we need to provide a thumbnail by using <video:thumbnail_loc>https://img.youtube.com/vi/<insert-youtube-video-id-here>/maxresdefault.jpg</video:thumbnail_loc>
And we need to provide the title which should match the page’s H1 tag and this should be wrapped in <video:title> </video:title>
And we need to provide the title which should match the description from the page (currently wrapped in itemprop=”description” tags on page) and be wrapped in <video:description> </video:description>
And we need to provide the content location which will be <video:content_loc>https://www.youtube.com/embed/<insert-youtube-video-id-here></video:content_loc>

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And that we have a lot of different types
And that when we list all of them the lists can become large
And that speed means a lot to us
And that aside from splitting up sitemaps we can also compress them
When the assets are published
Then we can publish gzipped sitemaps along site unzipped sitemaps
And provide these gzipped sitemaps to the search engines instead
And list these in a public sitemap index xml with the filename https://www.website.co.uk/sitemap.xml that we provide to the search engines
And this should be listed in robots.txt
And this should be the only one uploaded to search engine webmaster tools
And list the uncompressed in another file for internal SEO and human use that we publish at /sitemaps.xml

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And the search engines have a 50,000 item limit per sitemap
When the assets are published and added to an XML sitemap
Then we need to split any lists that go above this size
And if this happens it would make sense to split sitemaps into logical categories
But this should not need to be thought about for a while

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And that we can advise the search engines on when a page was last modified to help with crawl prioritisation
And that we can advise the search engines on what priority we rate a page from 0.0 (lowest) to 1.0 (highest)
And that we can advise the search engines on how frequently a page is likely to change (always[changes when accessed]/hourly/daily/weekly/monthly/yearly/never[archived pages])
When the assets are published and added to an XML sitemap
Then we need to add this data within the <url> </url> tags
And we need to use the <lastmod> tag in W3C datetime format with time omitted (e.g. 2017-12-25)
And we need to set a priority on pages
And that should be set to 1.0 for new arrivals, 0.9 for promo pages, 0.8 for all product pages and left to 0.5 (default) for all other pages (we can tweak later if required)
And we need to use the <changefreq> to advise on change frequency (always/hourly/daily/weekly/monthly/yearly/never)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we don’t want to have duplicate content being pushed to the search engines
When the assets are published and added to an XML sitemap
Then we need to list only canonical URLs
And we don’t want any pages that have a query string in them in the url to feature in the sitemap
And if we have important pages that have query strings in, these need to be changed to non-query based URL’s so the search engines don’t ignore them

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we have multiple country sites
And some we don’t provide certain services in some countries (e.g. serving)
When the assets are published and added to an XML sitemap
Then we need to have a specific set of sitemaps per country
And not list in the sitemap services we don’t provide in these countries
But we also need to make sure these pages do not have hreflang tags point them to non-existent pages within the pages themselves (added otherwise I will forget to tell you this)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And we have multiple country sites
And these are in multiple languages
When the assets are published and added to an XML sitemap
Then we need include hreflang alternatives for each page URL
But if this does not exist in the language or is not available as a service then we should not list it as an hreflang alternative on any sitemap (e.g. https://www.website.fr/service)

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And sometimes these pages are archived
When the assets are archived
Then we need to move them off all lists into an archive.xml
And give all these pages priority 0.0 And if this list exceedes 50,000 items we need to split the list and come up with a genius name like archive2.xml
And if their URL group contains any other assets they need to be included as well

Given that we have assets on the websites and that we want search engines to find them and index them quickly
And sometimes are removed
When the assets are unpublished or deleted
Then we need them to be removed from all sitemaps
And if their URL group contains any other assets they need to be removed as well

Reference documentation:

Microsoft 30 Digital Champions: Watchfinder Video Feature

Sat, 15 Oct 2016 09:00:00 +0000

Real Business and Microsoft have been running a series called “30 Digital Champions” - profiling UK companies that have used digital transformation to achieve significant growth. Watchfinder was selected as one of the 30, and I was asked to represent the company on camera.

The interview focused on how we’ve managed to double our stock and staff in the last 12 months while expanding internationally. It’s a question I get asked a lot, so it was useful to articulate it properly.

What we talked about

The conversation covered three main areas: our approach to technology investment, the move to cloud infrastructure, and how we’re using machine learning.

The “proof of worth” principle

One thing I was keen to explain is that we don’t invest in technology for its own sake. Everything needs to demonstrate ROI before we commit resources to it. A good example is organic social media. Despite what you read in marketing blogs, we’ve tracked exactly one sale that came directly or indirectly via Facebook. One. So we focus our attention on programmatic advertising instead, which has shown genuinely strong results.

This isn’t about being anti-innovation. It’s about being honest with ourselves about what actually drives the business forward.

Moving to Azure

We’ve recently migrated the entire company - front-end websites, back-office systems, the lot - onto Microsoft Azure. The main driver wasn’t cost (though that helps). It was security.

When you’re holding several million pounds worth of luxury watches, you’re an attractive target. That’s true physically, and it’s true online. By moving to Azure, we’ve shifted a lot of the security burden onto Microsoft’s teams, who are frankly better equipped to handle it than we are. We no longer deal with physical servers, biometrics, or patching. That frees our IT team to focus on productivity - which matters more now we’re running eight physical sites and three international websites.

Machine learning for pricing

The bit that seemed to interest the interviewer most was our work on machine learning. We’ve sold around 80,000 watches over 12 years and provided about 750,000 quotes. That’s a lot of data on what sells, at what price, and how quickly.

We’re now applying machine learning to automate pricing decisions. The goal is to reduce how much our salespeople need to be involved in individual pricing calls, while also improving the online experience for customers. It’s early days, but the potential is significant.

The filming experience

I’ll be honest - being interviewed on camera isn’t something I’m naturally comfortable with. You’re very aware of the lights, the multiple takes, and the knowledge that your words will be edited into something you can’t control.

That said, the Real Business team made it straightforward. They knew what story they wanted to tell and asked good questions to draw it out.

The video

Here’s the finished piece:

How to make Google algorithms work for you

Tue, 27 Jan 2015 07:30:59 +0000

Every second of every day, Google processes 40,000 search queries from all over the world. That’s more than 3.5 billion searches every year. With more than 100 million active websites hosted on the internet, how does the company sort through all this information?

It’s no secret that Google uses algorithms to match search results with queries. Algorithms are the computer processes and formulas used to match the search to the best possible webpages. Google’s algorithms rely on more than 200 unique signals or “clues” that make it possible to guess what the searcher wants. These signals include things like the keywords on websites, the freshness of content, location and authority.

When you conduct a search, the algorithm assigns a weighting to each web page, which determines where the website appears in the Google Search Engine Results page (SERP).

Search Features

As well as the established search engine results we’re all used to, Google regularly tries out new features to help users get the information they need more quickly:

Answers - Displays immediate answers and information for things such as the weather, sports scores and quick facts
Autocomplete - Predicts what you might be searching for
Freshness - Shows the latest news and information, including gathering timely results for specific search dates
Images - Shows you image-based results with thumbnails
News - Results from online newspapers and blogs from around the world
Video - Shows you video aggregated from various sources including YouTube, Vimeo and on-site players

Google’s technology is also fine-tuned to adjust searches from mobile devices such as tablets and smartphones, highlighting mobile friendly websites, along with advanced search options that allow the user to tailor the results to exclude unrelated fields and narrow the results returned.

Algorithm Updates

Around 500-600 times a year, Google makes minor changes to its algorithm. However, occasionally there is a major update that can affect search results in a big way. The purpose of Google’s frequent revision of its algorithm is to weed out the efforts of people who intentionally violate webmaster guidelines in an effort to get a higher ranking for poor quality sites.

The latest update, released October 2014, is Penguin 3.0. Just like its predecessors, Penguin 3.0 is designed to improve search results by eliminating spam or links that don’t appear to be authentic. Webmasters who use too many backlinks with the same anchor text and landing or doorway pages will need to revise or remove them to preserve their page ranking.

Another algorithm, Google Pigeon was launched in July 2014 and focuses on improving local search results. It uses location to return the most relevant results, which is especially good news for small businesses listed in the Yellow Pages or local directories. Google Hummingbird was released in 2013 and seeks to understand the query by incorporating synonyms and context in the results. It not only considers each word but also how the word was used in the query.

Ranking Factors

With the knowledge of how these algorithms work, there are several other factors that impact page rankings.

Content

There is a measurable correlation between the quality of content and rankings. Pages with poor quality or duplicated content should be removed, rewritten, or blocked from being indexed by the search engine. A keen webmaster will continue to update the content to prevent their pages from being downgraded in the search rankings. Experts suggest that longer content (over 500 words) trends higher in the page rankings.

On-Page Technical SEO

With on-page SEO, the keyword remains an important part of the overall concept. The keyword should be used cleverly in the title, as well as the description, body, and sub-heading. However, there is no need to write it in every sentence. Overuse of the keyword serves no purpose and should be avoided.

Backlinks

Backlinks are still important but the number of keyword backlinks continues to decrease. Big brands appear to benefit the most from the use of backlinks. Every backlink is like a vote, however Google takes into account the source of the link and adjusts the value, therefore a backlink from a high quality relevant source of good standing has much more value than a link from a directory for example.

Site Load

The amount of time it takes for your site to load is very important, especially since more people are accessing the internet from mobile devices.

User Behaviour

The click-through rate and time-on-site are usually higher in better ranking sites. These sites will also have a much lower bounce rate than lower ranking pages.

The Bottom Line

Google will continue to change its algorithms to adapt to the way we use the search engine. To maintain a high page ranking, the strategy for webmasters should be to produce great content and target the right keywords to match.

Even if you have done a poor job with SEO in the past, a keen understanding of the algorithms and focusing on improvements to the user experience are bound to work in your favour.

Seven Steps to Successful Content Marketing

Thu, 08 Jan 2015 07:30:59 +0000

As the competition in advertising heats up, more companies and major brands are turning to content marketing to attract customers and build awareness among consumers. Experts are predicting that search engine optimisation (SEO) and content marketing will go hand in hand to boost page rankings in Google search results.

Content marketing can take many forms and often viewers hardly realise that they are sponsored by major brands. It may be an entertaining article or an informative video. However, with so many avenues for distribution, the content can be developed in almost any format to be read or viewed on various devices. These include slideshows, marketing videos, images, press releases, events, webinars, games, apps, tools, e-books and whitepapers, email and e-learning.

The Seven Steps

Defining the content strategy
Choosing the best format
Content types that appeal to your target audience
Content distribution and social media
Key metrics and goals
Main sharing triggers
Double checking

Step 1: Define Your Strategy

The first step of content marketing is to develop a clear plan of the objectives you want to achieve. This process will include the topics to be covered, the audience to be targeted and the overall campaign goals. With consumer attention spans getting shorter, it might be wise to combine multiple formats into one piece of content. For example, a blog post with a professionally designed infographic will generate much more interest than a plain article.

Step 2: Choose Your Topics

The topics to be covered should also be clearly considered and relevant to the goals you wish to accomplish. However, stick to proven topics that work well, and that will appeal to the target audience and motivate them to share. Examples of popular topics include:

How to guides
Reviews
Q and As
Galleries
Compilations
Beginners guides
Checklists
Infographics

Step 3: Plan Your Distribution

You should also have a clear idea of how the content will be distributed. The options include hosting on a personal website or blog but this only works if the audience already knows about your website. Consider whether the topic is a good fit with other content on the site, and whether the host location is able to handle high traffic if the content goes viral. Social networking sites offer a mass avenue to distribute the content to millions of people at little or no cost.

Step 4: Measure Performance

After your content has been published, you need to be able to measure its performance. There are several ways to do this. Tools such as Google Analytics give access to highly valuable detailed information that will tell you if people are interested in your content and how they are engaging with it.

Basic information such as page views are further broken down into unique visitors, new visitors, and bounce rate to give a clearer picture of the true numbers versus spam and webmaster visits. You can also track behaviours such as downloads and sharing and use this to tweak the content or change your distribution channel. Using Google Analytics, you can also set goals that tie in to your primary business goals. Whether it is to generate traffic, drive sales, or to increase brand awareness, all content should relate to the overall goal.

From the outset, one of the main objectives should be to create good content that users will want to share over and over again. There are several emotional drivers that stir people to share content, whether it is funny, shocking, addresses a controversial issue, or inspires them on a personal level.

Step 6: Include a Call to Action

Don’t just share a generic topic. Include a call to action, such as an invitation for discussion or comments. Most importantly, follow any company or personal style guidelines and keep the overall campaign goal in mind.

Step 7: Review and Revise

The internet is not a place for static content. Review and revise your content periodically to keep it current and maintain interest over time. You can also offer the same topic in various formats to reach more people.

Summary

Successful content marketing requires companies and individuals to be strategic and deliberate with every step of the process. This carefully thought out process also involves keeping abreast of new developments in marketing as well as changes in audience behaviour.

Marketers should also be open to experimenting with different types of content with different distribution channels, and willing to adapt their methods to achieve the best results.

Day One Journal

Sat, 03 Jan 2015 21:23:59 +0000

So I have decided to take note of Robin Sharma’s excellent teachings and spend a little time journaling my life.

My guess is that given his general philosophy, that he would forgive me for using an app to journal rather than a written journal, as at least it means I am doing rather than not doing.

From Monday I start a new role as head of marketing for startup digital video production company Base Touch and I want to be able to journal my life a little in order to help organise it as well as remember.

Robin suggests writing your journal in the “Power Hour” between 05:00 - 06:00 however as this when I will be doing my morning workout or run, I will instead attempt to do a journal entry on my daily 1.5 hour commute from Chartham to London.

Many of my daily journals I will publish to my website, some I will keep private. edit: I have subsequently uesed this daily for personal journals only

So this is day 1 using Day One.

I chose Day One after finding it on the App Store looking for apps that supported markdown as I have just converted my personal website from Wordpress to Jekyll Static Site Generator (I intend to write a blog post on this move and my reasons for it within the next few days).

Further more I found that there is a Jekyll-DayOne plugin from Jeff Verkoeyen and frankly his blog looks excellent. But I don’t intend to set this up yet.

Anyway I am currently just trying out the Day One app on my iPhone whilst my wife, Nicole, has her first ever go of Minecraft with our two boys (massive Minecraft fans) something she promised to do for a while. Well done Nicole, you survived the first night!

When I have had a bit more of a play I will do a bit more of a run down of features and a review of Day One app.

– Header image by AJ Montpetit